Union College Union | Digital Works
Honors Theses Student Work
6-2011 The mpI act of the 2005 Collective Bargaining Agreement on Competitive Balance in the National Hockey League John L. Simpson Union College - Schenectady, NY
Follow this and additional works at: https://digitalworks.union.edu/theses Part of the Income Distribution Commons, and the Sports Studies Commons
Recommended Citation Simpson, John L., "The mpI act of the 2005 Collective Bargaining Agreement on Competitive Balance in the National Hockey League" (2011). Honors Theses. 1063. https://digitalworks.union.edu/theses/1063
This Open Access is brought to you for free and open access by the Student Work at Union | Digital Works. It has been accepted for inclusion in Honors Theses by an authorized administrator of Union | Digital Works. For more information, please contact [email protected].
The Impact of the 2005 Collective Bargaining Agreement
on Competitive Balance in the National Hockey League
by
John Simpson
*******************************
Submitted in partial fulfillment of the requirements for Honors in the Department of Economics
UNION COLLE GE June, 2011
Abstract
After a lockout that canceled the 2004‐05 season in the National Hockey
League (NHL), the owners and players reached a collective bargaining agreement
(CBA) that instituted a ‘hard’ salary cap, a modified revenue sharing system, and changes in free agency. The principal motivation for the new agreement was to raise competitiveness among the teams, in order to generate greater revenue and profitability and to support higher player salaries. The purpose of this study is to evaluate the impact of the CBA on competitive balance within the NHL and identify the principal determinants of the changes in competitiveness among the teams.
We evaluate several alternative measures of competitive balance: dispersion in point percentage; the Herfindahl‐Hirschman index; the Gini Index; number of playoff games; and number of times a team has made the playoffs. We then develop and estimate alternative models of the determinants of point percentage (equal to points earned by a team divided by total possible points), in which point percentage is hypothesized to be a function of team payroll (a proxy for player talent) and free agent signings.
We estimate the models using a sample of all 30 teams in the NHL across eleven seasons, including five seasons prior to the 2005 CBA (2000‐2004) and six seasons after its implementation (2006‐2011). We find that the new CBA has increased competitive balance, primarily because of a reduced impact of team payroll on point percentage and the facilitation of free agent signings which have more nearly equalized the talent across teams.
i
Table of Contents
List of Exhibits iii
Chaeptr One: Introduction 1 1. Importance of Study 2. Hypothesis 3. Outline of Study
Chapt er Two: Evolution of the Collective Bargaining Agreement 4 1. Concept of Competitive Balance 2. Payroll and Revenue Sharing Mechanisms 3. The 2005 NHL Collective Bargaining Agreement
Chapt er Three: Measures and Determinants of Competitive Balance 10 1. Background on Alternative Measures of Competitive Balance 2. Determinants of Competitive Balance (Model Statement)
Chapt er Four: Empirical Tests of Competitive Balance 23 1. Sources and Measurement of Data 2. Results from Measures of Competitive Balance 3. Regression Results for Determinants of Competitive Balance
Chaeptr Five: Conclusion 42 1. Interpretation of Findings 2. Policy Remarks 3. Suggestions for Further Research
Bibliography 47
Appendix A 50 Appendix B 52 Appendix C 56
ii List of Exhibits
Exhibit 1: Unrestricted Free Agency Rules 18
Exhibit 2: Calculation of Point Percentage 23
Exhibit 3: Descriptive Statistics 25
Exhibit 4: Standard Deviation and Team Point Percentage 26
Exhibit 5: Ratio of Standard Deviation to Ideal Standard 27
Exhibit 6: Herfindahl‐Hirschman Index 28
Exhibit 7: Distribution of Wins 29
Exhibit 8: Number of Playoff Games 30
Exhibit 9: H‐H Index with Playoff Appearances 32
Exhibit 10: Previous Point Percentage Measure 33
Exhibit 11: Chow Test for Exhibit 10 Regressions 34
Exhibit 12: Payroll Regressions 35
Exhibit 13: Chow Test for Equations 12.1 36
Exhibit 14: Model Statement Regressions 39
Exhibit 15: The Effect of Five Points on Playoff Teams 40
iii Chapter 1 Introduction
Purpose of the Study
In all professional sports, competitive balance is an important and desirable attribute to have within the League. All of the major American professional sports leagues are always looking for ways to increase competitive balance by changing the rules that govern their sport. Prior to 2004, the National Football League (NFL), the
Major League Baseball (MLB), and National Basketball League (NBA) all had payroll
mechanisms and revenue sharing plans to help increase its competitive balance.
Because the National Hockey League was the only major professional League that
did not have any of these provisions in its Collective Bargaining Agreement (CBA),
the resulting lack of competitive balance may have contributed to lower profits.
Competitive balance is important as it increases the quality of play between teams,
raises fan interest and attendance, and leads to more lucrative television deals, all of
which generate higher revenue and profitability. (Neale)
The purpose of this study is to evaluate the magnitude and causes of the
change in competitive balance within the League since the institution of the 2005
CBA, which added payroll mechanisms and a revenue sharing plan.
This research is also important as the current CBA is set to expire at the end
of the 2011‐12 season. Without a doubt, the League owners and players union will
sit down together to discuss the state of the NHL. In doing so, one of the concepts
that will be discussed is the competitive balance of the League. They will look to see
if the changes they made positively or negatively affected the game and the profits
of the League. If there is a positive effect, then there is a chance they could include
1 more payroll mechanisms or a different revenue sharing plan to further increase the
effects on competitive balance. If negative effects are found, then they will know that a different system is necessary.
Outline of the Study
In the second chapter we will look into the institutional and historical context of competitive balance. We will introduce various ways payroll can be
controlled by the League, including different salary caps and salary floors. We will also indulge into revenue sharing plans and their purpose of spreading League wealth. Finally, we will describe the important changes in the 2005 NHL Collective
Bargaining Agreement.
In the third chapter we first examine alternative measures of competitive
balance, including standard deviation of point percentage, a Herfindahl‐Hirschman
Index, distribution of wins, number of playoff games, number of times a team makes the playoffs, and a comparison to the previous season’s point percentage. We then will examine the determinants of competitive balance, including the institution of the salary cap, revenue sharing, and changes in free agency. We will end with the introduction of the simple model, using point percentage as the dependent variable and payroll and free agent signings as the dependent variables.
In the fourth chapter, we introduce the sources of the sample. Next, we estimate and analyze the various measures of competitive balance. Lastly, we estimate the model and review various regression results.
2 In the final chapter, we conclude that competitive balance has improved as a result of the 2005 CBA. Also, we find that the impact of the changes in the CBA equalized the capacity of teams to acquire players of talent, which were the driving forces in the improvement of the League’s competitive balance.
3
Chapter 2 Evolution of the Collective Bargaining Agreement
In this chapter, we begin by introducing various ways payroll can be
controlled by the League, including different salary caps and salary floors. We then will introduce revenue sharing plans and their purpose of spreading League wealth.
Finally, we will introduce the 2005 NHL Collective Bargaining Agreement and detail the important changes.
Concepts of Competitive Balance
Competitive balance in any market is described as a market situation where
no business is too big or has an unfair advantage. (Financial Times Lexicon) In
sports, it can be much different. There are many ways in which competitive balance
has been studied. In general, competitive balance shows the uncertainty in sporting
events.
In professional sports leagues, there are two types of competitive imbalance.
(Sanderson and Siegfried, 2003) One is where the large market teams with lots of money acquire all of the best players. This is the most common form of imbalance and the one that this paper addresses directly. There are many methods that can be used to control competitive balance. Some of the methods that have been used in professional sports leagues are payroll mechanisms and revenue sharing plans.
4 Ways to Limit the Size of Payroll
Payroll mechanisms are used to place boundaries on the amount of money
that a team can spend on payroll. A ‘salary cap’ is an example of one of these mechanisms. This prevents the richer teams from over‐paying players to get them to come and play for them. There are two types of salary caps, ‘hard’ caps and ‘soft’ caps. A hard cap is what the NFL uses.1 With a hard cap, team payroll cannot
exceed the limit for any reason whatsoever. This cap is usually calculated as a
percentage of league revenue. The soft cap is different in that teams can exceed the
limit under certain circumstances. In the NBA, a team can exceed the cap in order to
resign a player who is already on the team. (Castillo, 2010) This is done to prevent
big name players from changing teams a lot, which could lead to fans becoming
disinterested. This cap still prevents teams from going into free agency to sign the
best players. The MLB also uses a type of a soft cap called a luxury tax.2 With this
system, teams are taxed a certain percentage for every dollar they go above the cap.
With this system, richer teams are still able to acquire the better players, as long as
they are willing to pay the tax.
Another payroll mechanism similar to a salary cap is a salary floor. This is a
lower limit of team salary that every team must meet. The idea here is to force
teams to spend money on players to become more competitive. There are pros and
cons to using a salary floor. One pro is that it forces owners to sign players, which
will help them to improve. The major con is that it can force a team to overpay for
1http://images.nflplayers.com/mediaResources/files/PDFs/General/NFL%20COLL ECTIVE%20BARGAINING%20AGREEMENT%202006%20‐%202012.pdf 2 http://mlbplayers.mlb.com/pa/pdf/cba_english.pdf
5 mediocre players, which will not help to increase the competitiveness of the team.
They will not get the best performance for the amount of money they are spending.
Thus, the team will still be uncompetitive.
Revenue Sharing
A League can also use revenue sharing techniques to increase competitive balance. The idea behind revenue sharing is to give all teams an equal chance to retain and sign players. This will help the poorer teams to sign expensive players.
There are many ways a League can institute revenue sharing. One way is to have every team contribute a percentage of its revenues into a pot each season. This pot will then be equally distributed among all of the teams. Under this system, a rich team will pay more to the pool, but will receive the same amount as a poorer team.
This system is used in the MLB.3 The strongest revenue sharing plan is the one used in the NFL.4 This is similar to the MLB in that all the League revenue is split evenly among all the teams. The NFL also uses a gate revenue sharing system. (Szymanski and Kessenne, 2004) In this situation, a certain percentage of revenues from games are put into a pool. This pool is then also equally divided amongst teams. Since the
NFL makes so much money from broadcasting games on national television, the poorer teams receive more than enough money needed to build and support a strong franchise.
3 http://mlbplayers.mlb.com/pa/pdf/cba_english.pdf 4http://images.nflplayers.com/mediaResources/files/PDFs/General/NFL%20COLL ECTIVE%20BARGAINING%20AGREEMENT%202006%20‐%202012.pdf
6 Although it is still tough to stop richer teams from getting better players, all
of the above mentioned plans could work to help make the poorer teams more
competitive. This is something that the NHL was missing out on before its new CBA
in 2005.
The 2005 NHL Collective Bargaining Agreement
Before the 2005 Collective Bargaining Agreement, the NHL was the only
professional sports league of the four major ones that had no salary cap, salary floor, or revenue sharing system. One of the main causes of the 2004‐2005 lockout was the escalating player salaries. The League wanted a salary cap that was linked to league revenues. This would help to stabilize league profits, and to increase the competitive balance of the League. The League had lost $273 million in the 2002‐
2003 season. (Bissonnette, 2007) The lack of salary cap and revenue sharing made it hard for poorer teams to compete with the richer teams. Before the 2004‐2005 season started, a new CBA was needed. The League and the National Hockey League
Player’s Association (NHLPA) went back and forth with offers. The League wanted a
hard salary cap and revenue sharing similar to MLS. The NHLPA denied these offers
and instead proposed a system with luxury taxes, a rollback in player salaries, and a
change to the League’s entry‐level system. The two sides were not close to coming
to a conclusion, which caused the 2004‐2005 lockout. Negotiations between the
two sides carried on for months and it did not look like an end was in sight. The
lockout was resolved when the League agreed to put in a revenue sharing plan, in
which the top ten moneymaking clubs must put money in a pool to be distributed to
7 the bottom fifteen teams (in terms of revenue). These bottom teams have to have a
market with 2.5 million television households or less, in order to be eligible for
revenue sharing.
As a result of the agreement, the NHLPA agreed to a hard salary cap that
would be tied to League revenues. No teams were allowed to exceed the salary cap
for any reason other than to replace a long‐term injury. For the 2005‐2006 season,
the cap was set at $39 million. League revenues have increased every year since the
lockout, leading to an increase in the salary cap as well. In the most recently
completed season, 2009‐2010, the salary cap was $56.8 million.
Along with the salary cap, the NHL also decided to put in a salary floor. No
team is permitted to have a team payroll below the salary floor. Originally, the plan
was to have the floor set to 55% of the salary cap. Over time, with an increase in revenues, the League decided to have it set at $16 million below the cap. In the
2009‐2010 season, the salary floor was $40.8 million.
Under the new CBA, there were also big changes in free agency; most notably players become unrestricted free agents. Each of the four years following the new
CBA, the unrestricted free agents got younger. The minimum age at which a player
can shop himself on the open market was 31 years old in 2005, 29 years old in 2006,
28 years old in 2007, and 27 years old in 2008 and beyond.5 This has allowed teams
to go after players at a younger age, when they are more valuable and can help make
a team become more competitive.
5 All information on the 2005 CBA was found at two sources: http://www.nhl.com/ice/page.htm?id=26366 http://www.nhl.com/cba/2005‐CBA.pdf
8 With the changes, the League was hoping to control player salaries, and therefore costs. They were also looking to increase competitive balance in order to increase revenues and profits.
In this study, we will see if competitive balance in the NHL has changed since the institution of the 2005 CBA. Furthermore, we will analyze the causes of the change in competitive balance by looking at the institution of the salary restrictions, revenue sharing plan, and free agency rule changes.
9 Chapter 3 Measures and Determinants of Competitive Balance
In this chapter, we will introduce the various measure and determinants of
competitive balance that are used in this study. Alternative measures include a
standard deviation of point percentage, a Herfindahl‐Hirschman Index, a Gini‐ Index,
distribution of wins, number of playoff games, number of times a team makes the
playoffs, and a comparison to the previous season point percentage. The
determinants that we will focus on include the institution of the salary cap, revenue
sharing, and changes in free agency. We will end with a simple model statement,
which uses point percentage as the dependent variable and payroll and free agent
signings as the independent variables.
Defining Point Percentage
The most often used measure of competitive balance is winning percentage.
This is because it is one of the best measures of a team’s success. If teams have a
high winning percentage, they are doing well in the standings. We will use a similar
measure, with a small difference. We will use the point percentage of a team as a
measure. In a competition, a team gets 2 points for a win, 1 point for an overtime
loss, tie, or shoot‐out loss, and 0 points for a loss in regulation. The point percentage
is calculated as follows.
Point Percentage = Total Number of Points Total Possible Points
For any team, the total possible points is equal to 2*n, where n is the number of
games played. The reason this method is used is because losing a game while still
10 getting a point will affect placement in the standings. Thus, it is important to
include these points in the measures. Also, an overtime loss or shootout loss shows
that the game was only decided by one goal, meaning it was a competitive match. A
loss in regulation can be by much more, and shows less competition in the match.
For these reasons, point percentage will be used as a measure of team performance.
Alternative Measures of Competitive Balance
Spread of Point Percentage
One measure is looking at the standard deviation of team point percentage.
The further spread out the data is, the less competitive the League is. For perfect
competitive balance, this number would be zero. This is just a basic measure that
will only be able to tell part of the story.
Point Percentage Compared to an Ideal Standard
Another similar measure compares team‐point percentage to an ideal
standard. This is done in previous works that have measured competitive balance.
(Richardson, 2000 and Leeds and Allmen, 2002) In each game, under perfect
competiveness, a team has a probability of .5 of winning. Thus, the ideal measure is
equal to .5/sqrt(n) where n is the number of games each teams plays. (Leeds and
Allmen, 2002) In the NHL 82 games are played, so this ideal measure of point
percentage standard deviation would be .5/sqrt(n) = .5/sqrt(82) = .055216. So, in
order to use the measure, each team’s point percentage will be divided by this ideal
11 standard. Then, the standard deviation of these will be taken. Once we have this,
we can use the formula below.
Ratio = S D of team point‐percentage ideal standard
This is a good measure to use because it will show how spread out the teams are
from an ‘ideal’ spread of point percentage.
There is another measure of competitive balance using a different ideal
standard. This measure accounts for ties. Before the NHL lockout, if the game
ended in a tie after overtime, then the game would be recorded as a tie. After the
lockout, the game would go to a shoot‐out to determine the winner. This causes a
difference in point percentage from the previous ‘ideal’ standard. This measure
does so by looking at the probability of a tie. Under equal playing strength, this
standard deviation is equal to (1‐p/4n)^1/2 where n is the number of teams and p
is the probability of a tie. (Richardson, 2000) Since ties have been eliminated as a
result of the new CBA, this may cause a problem when comparing the data before
and after the lockout. Also, there is no set probability for a tie. This means data
from previous seasons must be used to see what the historical probability has been.
If seven years of data previous to the lockout are used, the probability of a tie is
13.41%.6 This would make the equation become (1‐p/4n)^1/2 =
(1.1341/4(82))^1/2 = .0514. After the lockout, this value becomes (1‐p/4n)^1/2 =
(1/4(82))^1/2 = .055216. Each team’s point percentage will be divided by this value, and then the standard deviation will be taken.
6 http://www.nhl.com/ice/standings.htm?season=20032004. There is a drop down menu to scroll through previous seasons.
12
Using both of these different ideal standards, we will see if the change in
competitive balance depends on the ‘ideal’ measure that is used.
HerfindahlHirschman Index
Another way to measure competitive balance is to consider the distribution of the number of wins, rather than point percentage. The Herfindahl‐Hirschman (H‐
H) Index measures inequality in match and championship outcome and is given by:
2 HHI= Σ(MSi) where MS is the market share of the ith firm. In this case, the market share for a
team is its number of wins divided by the total number of wins by all teams:
2 HHI = Σ(Wit/ΣWit)
where W is the number of wins by the ith team in the tth year . (Owen, Ryan, and
Weatherson, 2006) The closer that this measure is to 0, the closer the League is to perfect equality.
GiniIndex/ Quintile Approach
Another measure uses the Gini‐index. This method was previously used to look at the effect that the institution of a hard salary cap had on competitive balance in the NFL. (Sommers et al, 2004) This measure is used by ranking teams in order of number of wins. Then, we do the following calculation;
Gini index = ∑ (fi*yi+1 − fi+1*yi),
where f is the cumulative percentage of teams and y is the cumulative percentage of
13 wins. In this study, we will use this measure, but with a small change. We will use
the Quintile approach where teams will be placed into 5 equal‐sixed groups. In a perfectly competitively balanced League, each group will have 20% of the total wins.
This measure will be tested across each season to see if there is a more uniform distribution across quintiles in the seasons after the lockout.
Playoff Measures of Competitive Balance
Competitive balance can also be evaluated based on playoff performance.
Although the regular season is important, most managers and players are more
concerned about the playoffs. For managers, it is a chance to increase revenues and
fan interest. For players, it is a chance to prove that they are the best at what they do. It is almost like another season being played.
Number of Playoff Games
A simple measure of the competitive balance is the number of playoff games that take place in a season. (Richardson, 2000) Every season, eight teams from the
Eastern Conference and eight teams from the Western Conference make the playoffs. The teams are ranked based on points earned in the regular season. After they are seeded, the first seed plays the eighth seed, the second plays the seventh, and so on. This is done in each conference. The teams then play a best of seven series with the first team to four wins moving on to the next round. The more competitive a series, the more games it will take to decide a winner. The winners of each series move on and are re‐seeded based on the teams that remain. There are a
14 total of 15 series in the playoffs. This means there can be anywhere from 60 to 105
games. The limitation of this measure is that it does not take account of the
closeness of the games. If all of the games are won by one goal, then the series is
more competitive than if the games were all blowouts. This will not be measured in
here.
Playoff Appearances
Another measure is team shares of playoff appearances across seasons. In a perfectly balanced League, every team would make the playoffs approximately
every 2 years. The formula for this Index is;
2 HHI = Σ(Pi)
where P is the percentage of seasons that team i has qualified for the playoffs.
(Owen, Ryan, and Weatherson, 2006) As this index increases, competition
decreases. The value of this index is that it looks at what every team is most
concerned about, getting to the playoffs. If some teams continually make the
playoffs, more frequently than others, then this will result in higher squared
numbers. The only limitation in this measure is it does not really tell the whole
story. Even if the same teams are making the playoffs, their rankings within the
playoffs can be different from year to year. This is an aspect of competitive balance
that is not measured in this index.
15 Previous Year’s Point Percentage
The last measure that will be used shows the effect that last season’s point percentage has on this year’s. A simple regression model can be used:
PointPct = α + β1PointPct‐1 + ε,
where PointPct is the point percentage of the current year and PointPct‐1 is the point
percentage from last year. This regression will be run across two different time
periods, both before and after the lockout. In comparing the two time periods, we
can see if the 2005 CBA reduced the effect, suggesting an increase in competitive
balance.
Now that we have established the ways of which we will measure
competitive balance, we next look at its determinants of team performance.
Determinants of Team Performance
Payroll
With the salary cap, teams will not be able to go out and spend all the money
that they want. This will bring team payrolls closer together. Teams closer to the
cap will not be able to sign players seeking a lot of money, while the teams with lots
of ‘cap room’ will be able to sign those players. So, the teams in the League should
have payrolls that are closer to each other. Team payroll will be compared to team
point percentage to see if they share a relationship. Previous studies have looked at
the relationship between team payroll and point percentage. Forrest and Simmons
(2000) look at all four major sports in America to see this effect. It looks at team
16 payroll to see if it is a significant predictor of team success, when measured by point
percentage in the regular season. It found that it was a strong predictor for all four
major US sports. This should help to explain the changes in competitive balance as a
result of the new CBA. Berri and Todd (2004) look at the NBA to see if team payroll affects the number of wins. They found that salary dispersion has no effect on the number of team wins. This suggests that teams are not getting what they pay for, and there are other determinants that are more important. Sommers et al. (2004) look at the effects of a hard salary cap in the NFL. This salary cap brought team payrolls closer together, but this had no effect on performance, measured by the
Gini‐index.
This study will experiment with the effect that payroll has on team performance. It could very well be the case that this effect is not linear, where every dollar spent has an equal effect on performance. Logarithmic and quadratic specifications will be considered to capture the true effect that payroll has. These relationships will allow us to see if there are diminishing returns to payroll on performance.
Free Agency
Next, the changes in free agency rules will be looked at to see its effect on point percentage. As noted in Chapter 2, the major changes to unrestricted free agency are as follows;
17 Exhibit 1: Unrestricted Free Agency Rules Minimum Age of Year Unrestricted Free Agents 2005 and before 31 2006-2007 28 2007-2008 28 2008-2009 and 27 after
With more players becoming unrestricted free agents, there will be more opportunities for weaker teams to pick up players that will help them be more competitive. Before this change, the players that were available were older players who do not have the same impact on team performance as the younger players. We hypothesize that the number of free agents signed by a team does have an effect on team point percentage. Although not much research has looked directly at this, there have been multiple papers on Free Agency and its effect in general. Fishman
(2003) looks at free agency in the MLB. He uses a few different measures to see the effect and they show conflicting results. In general, Fishman finds that the higher the number of free agents available, the greater the effect on team performance.
This has shown to help the teams that are willing to spend the money on the free agents. Other research has shown different results. Fort and Lee (2007) look at the other three major sports in America to see the effects of free agency. They find that there is no effect of free agency on the competitive balance of a League.
Free agency is also affected indirectly with a change in the CBA. This change has to do with the salary cap. It has made it harder for teams to resign players when they are up for a pay raise. If the team does not have the room, they must let them go. Thus, there will be more free agents available for teams to pursue. For both of
18 these reasons, weaker teams should be able to improve and become more competitive, while performance of better teams is reduced.
It is also important to note that the type of free agent signed will have different effects. In any given year, numerous free agents are available who will bring different skills to a team. Some of these players are better than others. As a result, these free agents will need to be split into groups. This is because the effect of signing two of the best players is much different from signing two of the worst.
For this study, we have broken the free agents into three groups, top‐end, medium‐ end, and low‐end players. The high‐end players will most likely improve a team’s chances while low‐end players may not even dress every game of the season.
Revenue Sharing
Finally, the addition of revenue sharing in the NHL is an important determinant of competitive balance. Revenue sharing is supposed to work by giving the poorer teams more money so that they can compete with the richer teams, when it comes to holding on to players and signing new players. This is the determinant of competitive balance that has been studied the most, and has also showed conflicting results. Some previous studies have shown revenue sharing having no effect or even a negative effect. One reason for this is because the team that receives the money does not have to spend it. If managers and owners are profit maximizing, then they may pocket the money that they receive from revenue sharing instead of investing it to increase the competitiveness of the team. Chang and Sanders (2009) found that revenue sharing led to an increase in the variance of expected winning
19 percentage. This is because sometimes teams in the lower half are not spending the
money. As a result, the better players will go to the better teams. Vrooman (2009)
argues that in order for revenue sharing to be effective, certain condition must be met. First, the League should share at least 50% of its revenue. He argues that this will lead to an increase in competitive balance. The NFL is a prime example. It shares all of its League revenue so all the teams have equal opportunity to build a winning program. Also, the owner must be more concerned with maximizing wins
rather than maximizing profits so that the money received from revenue sharing
will actually be used to improve the team.
Regression Model
Payroll and free agent signings will be used in the regression model, while revenue sharing will not be included. For one, there is not a good measure of revenue sharing that can be put into a regression as an independent variable. If the regression finds that payroll does have a lower effect in the seasons after the lockout, then it is reasonable to conclude that more revenue sharing will lead to more competitive balance. So, this regression will use payroll and number of free agents signed by a team as the determinants of point percentage.
The regression model is;
PointPct = α + β1Payroll + β2TFA1 + β3TFA2 + β4TFA3
Where PointPct = Point percentage Payroll = Payroll TFA1 = top‐end free agent TFA2 = medium‐end free agent TFA3 = low end free agent
20 Point percentage is used as the dependent variable because it is the most
appropriate measure of team performance. This regression model will be used with
two different time periods, before and after the lockout. This will allow is to
compare the results to see if there is a lowered effect of payroll on a team’s
performance. If this is the case, then we can say that the restrictions on payroll are a
reason for the change in competitive balance. This model will also be able to
determine the effect that signing a free agency has on point percentage. If this is
positive, then we can say that changes in free agency rules is a reason for the change
in competitive balance. In this model, we expect Payroll, TFA1, and TFA2 to be
positiv e, while TFA3 to be negative.
This paper will also consider logarithmic and quadratic specifications with
regards to payroll. The following two models regression models wil l be used.
2 Exponential ‐ PointPct = α + β1Payroll + β2Payroll Logarithmic ‐ PointPct = α + β1LogPayroll
In considering these two models with the first one given, we should be able to
capture the true effect that payroll has on a team’s performance.
With these regressions, we expect to find that payroll has a lower effect after
the lockout. This means that teams are not able to “buy championships” by
spending more money on players than other teams. We also expect to find that signing free agents will increase a team’s point percentage.
21 Chapter 4 Evaluation of Competitive Balance and Determinants of Team Performance
In this chapter, we will conduct the necessary estimates of the measures and determinants of competitive balance discussed in chapter 3. We first will introduce the sources of the sample. Next, each of the measures of competitive balance will be estimated and analyzed. Finally, the regression results estimating the model statement will be presented and analyzed.
Sources and Measurement of Data
Data comes from the 1999‐2000 season to the 2009‐2010 season plus part of
the 2010‐11 season (through February 18th). The 2004‐05 season is not used
because of the lockout. So, the data will cover the five years before and five and a
half years following the lockout. For each season, the data will cover the 30 teams in
the NHL. The one exception is the 1999‐00 season. In this season there were only
28 teams. In the following season, the Columbus Blue Jackets and the Minnesota
Wild entered the League as expansion teams.
The necessary data for each team includes; point percentage, wins, number of playoff appearances, number of playoff games, payroll, and number of free agents signed (by group). The data are collected across all 11 seasons.
To illustrate how the data set was derivied, a team, the Colorado Avalanche, will be used to walk through the steps to getting the data. First, the data that is necessary for the measure of competitive balance section will be introduced. This data includes point percentage, wins, number of playoff appearances, and the
22 number of playoff games. In order to get Colorado’s point percentage, the NHL’s
main website was used.7 An example of the calculation of point percentage is
shown in Exhibit 2.
Exhibit 2: Calculation of Point Percentage From the 2009-2010 Seasons Wins in Regulation: 43 Losses in Regulation: 30 Over Time Losses: 9
Point Percentage = Total Points , Total Possible Points
= (2*Wins) + (Ties + Overtime Losses) 2* Games Played
= (2*43) + (30 + 9) 2*80
= .579
The same calculation was use to calculate the point percentage for the other seasons
as well. In order to see if Colorado made the playoffs in a given year, this same website is used. In ranks the teams in order of points, and splits up the teams that made the playoffs from the teams that did not. We just looked at the standings for each year to see if Colorado was in the playoffs. To find the number of playoff games, the Hockey Nut website was used.8 This site shows the number of games for all NHL playoffs dating back to 1996‐1997.
7 http://www.nhl.com/ice/standings.htm#?navid=nav‐stn‐main (drop down menu to get to previous seasons) 8 http://www.hockeynut.com/archive.html
23 In order to find payroll information for the Colorado Avalanche, multiple
sources of data are necessary. One source is NHLnumbers.9 This gives Colorado’s
team payroll for the 2007‐08 season to the 2011‐2012 season. In order to get the
data in early seasons, Hockey Zone Plus was used.10 This gives Colorado’s payroll
information back to the 1992‐1993 season. From these sites, all the payroll
information is available. Finally, two sources were necessary to get information on
free agent signings. One of the sources is Sports City.11 This gives data on the
number of free agents that Colorado signed in the 2008‐2009 season and on. The
other seasons are received at Pro Ice Hockey, which gives data for the remaining
years.12 Next, it is necessary to look up each free agent’s salary to determine if they
are a top, mid, or low‐end free agent signing. There is one problem with the free
agent signings. Information on the free agent signings in the seasons prior to the
2004‐2005 season is unavailable. As a result, these seasons are not included in
some of the regressions later in the chapter.
Although steps are only shown for one team, all other teams are done the same way. The descriptive statistics of the data are reported in Exhibit 3.
9 http://nhlnumbers.com/teams/COL?year=2011 10 http://www.hockeyzoneplus.com/$maseq_e.htm 11 http://www.sportscity.com/nhl/2009‐nhl‐unrestricted‐free‐agents‐by‐position 12 http://proicehockey.about.com/od/nhlfreeagents/a/06_transactions_3.htm
24 Exhibit 3: Descriptive Statistics
00 01 02 03 04 06 07 08 09 10 11 Payroll ($millions) Average 31.62 33.34 38.03 42.38 43.95 34.32 40.30 44.37 53.40 54.34 56.54 Standard Deviation 9.96 10.36 12.58 13.59 15.46 6.38 4.72 7.58 4.87 4.97 5.69 Range 42.80 40.20 46.20 48.70 54.60 26.00 23.30 33.30 15.59 17.78 19.77 Point Percentage (%) Average 0.5249 0.5249 0.5246 0.5315 0.5295 0.5572 0.557 0.5553 0.5573 0.5613 0.5575 Standard Deviation 0.10 0.11 0.09 0.09 0.10 0.10 0.10 0.06 0.08 0.08 0.09 Range 0.46 0.40 0.38 0.32 0.31 0.41 0.35 0.27 0.34 0.36 0.37 Number of Wins Average 35.64 35.93 36.03 35.77 35.33 41.00 40.67 41.00 41.00 41.00 22.73 Standard Deviation 8.76 9.09 7.83 7.96 8.19 8.96 8.86 5.40 7.18 6.99 4.36 Range 37.00 31.00 32.00 30.00 28.00 37.00 31.00 23.00 29.00 27.00 16.00 Top Free Agents Signed (# of players) Average 0.60 0.47 0.47 0.50 0.53 .51 Standard Deviation 0.62 0.68 0.68 0.63 0.68 .64 Range 2.00 2.00 2.00 2.00 3.00 3.00 Mid Free Agents Signed (# of players) Average 1.53 1.73 1.33 1.37 1.30 1.35 Standard Deviation 1.14 0.94 1.03 0.96 1.06 1.04 Range 4.00 5.00 3.00 3.00 4.00 3.00 Low Free Agents Signed (# of players) Average 1.37 1.53 1.00 2.47 2.37 2.26 Standard Deviation 1.07 1.07 1.14 1.36 1.73 1.28 Range 3.00 4.00 5.00 6.00 6.00 5.00
25 Alternative Measure of Competitive Balance
Standard Deviation of Point Percentage
The first measure of Competitive Balance is the standard deviation of the point percentage across all of the seasons. The results are listed in Exhibit 4.
Exhibit 4: Stardard Deviation and Team Point Percentage
0.12 0.101 0.1 0.104 0.108 0.098 0.08 0.091 0.093 0.095 0.085 0.085 0.079 Deviation 0.06 0.064 0.04 Standard 0.02
0 2000 2001 2002 2003 2004 2006 2007 2008 2009 2010 2011 Season
Exhibit 3 shows that the standard deviation of point percentage is steady around .1 until the lockout. After the lockout, the results seemed to be the same for a couple years, before a dramatic drop in the 2007‐2008 season. In the past two seasons, the standard deviation has also been lower than before, ending at around .8 in the 2010‐
2011 season.
26 Point Percentage Compared To Ideal Standard
The next two measures compare the standard deviation of point percentage to an ideal standard. They are labeled “ideal1” and “ideal2”. The difference between the two is that ideal2 includes the probability of a tie in its measure of an ideal standard. These measures divided team point percentage by the ideal standard, and then the standard deviation is taken. The results from both of these methods are shown in Exhibit 5.
Exhibit 5: Ratio of Standard Deviation to "Ideal" Standard 2.5
2 2.02 2.10 1.86 1.82 1.5 1.78 1.78 1.53 1.55
Ratio 1 1.15 Ideal1 Ideal2 0.5
0
Season
Both of these measures that compare point percentage to an ideal standard provide
similar results. They both start right around 2, which is the ratio of actual to ideal
standard deviation. In the years up to the lockout, the standard deviation was
pretty steady anywhere between 1.7 and 2 for both measures. After the lockout,
27 when the measures become the same, we can see results that are similar to Exhibit
3. There is a two‐year period before the effect kicks in. After the dramatic decrease in the 2007‐08 season, it comes back up to right around 1.5.
HerfindahlHirschman Index
Next, the Herfindahl‐Hirschman (H‐H) measure is used. This measure is given by
2 th th HHI = Σ(Wit/ΣWit) , where W is the number of wins by the i team in the t year.
This is compared to the total number of wins by all teams. The results are given in
Exhibit 6.
Exhibit 6:HerfindahlHirschman Index 0.036
0.0355
0.0354 0.035 0.0349
Deviation 0.0348 0.0348 0.0348 0.0345
0.034 0.0343 0.0343 0.0342
Standard 0.0338 0.0335
0.033
Season
The results show that over time, the H‐H index has decreased from about .0355 to about .0342 over the ten and a half season spread. Again, this graph shows that there was a two‐year layover from the lockout period to where there is a noticeable difference in the results.
28
Distribution of Wins: Quintile Approach
Next, we evaluate the distribution of wins to determine the change in
competitive balance. This measures groups the teams based on number of wins and
uses the quintile approach to measure the percentage of wins% that each group
accounts for. The results are given in Exhibit 7.
Exhibit 7: Distribution of Wins Percentage of Teams (in order of number of wins) 0-20% 21-40% 41-60% 61-80% 81-100%
2000-01 0.129 0.166 0.204 0.23 0.27
2001-02 0.133 0.18 0.208 0.229 0.25
2002-03 0.144 0.167 0.194 0.231 0.263
2003-04 0.132 0.166 0.215 0.233 0.253
2004-05 NA NA NA NA NA % of Wins 2005-06 0.132 0.187 0.206 0.218 0.257
2006-07 0.134 0.176 0.208 0.233 0.249
2007-08 0.162 0.187 0.202 0.213 0.236
2008-09 0.15 0.184 0.201 0.218 0.247
2009-10 0.154 0.181 0.197 0.222 0.247
2010-11 0.153 0.183 0.206 0.216 0.243
This chart shows how the wins are spread out across teams in the NHL. In the
seasons before the lockout, the bottom 20% of teams have a small percentage of the
total number of wins. It maxes out at 14.4%, but on average is .135. Two years after
the lockout, the percentage jumps to 16.2% and does not fall below .15 for the
following three seasons. There is also a change in the top 20% of teams. Before the
29 lockout, the top 20% of teams shared anywhere between 25% and 27% of the wins.
Two years after the lockout, this proportion drops down to 23.6% and does not rise
above 24.7% for the next three seasons. There is not much change made to the win distribution of teams in the middle, but not much movement is expected. This is because the changes that are made in the 2005 CBA affect the top and bottom teams, but do not effect the middle teams as much.
Number of Playoff Games
The next method deals with the number of playoff games in a season. The measure works by adding the number of games in a playoff year. The results are listed in Exhibit 8.
Exhibit 8:Number of Playoff Games 92 90 90 88 89 89 89 86
Games 87 86 of 84 85 82 83 83
Number 80 81 78 76
Season
The number of games is maxed at 90 in the 2001‐02 season and minimized at 81 in
the 2006‐07 season. This is different from previous findings, which shows
30 improvements in competitive balance. After the lockout, the average number of
playoff games is lower than before the lockout, 85 and 87.4 respectively. The
number of play‐off games has increased each year since the 2006‐07 season.
However, it cannot be certain that this will continue into following seasons. These
results can suggest one of two things. For one, the CBA did not have an effect on the
competitive balance of the League. Second, it could be that the League has no
trouble with competitive balance in the playoffs. Since the playoffs only include the
best teams in the NHL, it has always been competitive. The problem with the balance occurred with the bottom half of the League, or the difference between the top half and the bottom half. If this is the case, then this method does not measure competitive balance in the way that the League wants it.
Playoff Appearances
Next, we will look at the number of times that a team has made the playoffs. For this measure, the Herfindahl‐Hirschman Index method is used. This measure is given by
2 HHI = Σ(Pi) , where P is the number of playoff appearances by team i in a five‐year
span. The results are listed in Exhibit 9.
31 Exhibit 9: HH Index with Playoff Apearances 320
310
300
290 appearances)
of 280 #
(in 270
260 Index
H
250
H 265 298 308 240 2006‐2010 2000‐2004 1995‐1999 Years
Exhibit 9 shows that the numbers are much lower after the lockout. With this measure, the higher the number, the less competition there is amongst teams to make the playoffs. These results show us that, before the lockout, the same teams would be making the playoffs every year. Now, the number of times a team makes the playoffs is more spread out. This suggests that the League is getting more competitive as different teams are making the playoffs each year.
Previous Point Percentage Measure
The last measure is the impact of a team’s point percentage from the prior year on the current year’s performance. The results from the regression are given in
Exhibit 10.
32 Exhibit 10: Previous Point Percentage Measure Dependent Variable: PointPct
Mean Sample Dependent 2 Size Variable C PointPct-1 R Eq. 145 0.5305 .216 .602 10.1 (6.48) (9.57) 0.391
Eq. 150 0.5579 .297 10.2 (7.79) .472 (6.9) 0.244 Source: Appendix A
These regression results show the effects of the variable PointPct‐1 for the seasons before and after the lockout. Before the lockout, the effect that PointPct1 had on the current point percentage (.602) is greater than after the lockout (.457). The regression before the lockout also has a higher R2 value, .391 compared to .236. In order to see if the difference is statistically significant, a Chow test is used.
In order to run a Chow test, we need a third regression that runs across both
time periods. After this, we conduct the following calculation;
where Sc = Sum of squared residiuals from full time period S1 = Sum of squared residuals from period before lockout S2 = Sum of squared residuals from period before lockout N1 and N2 = Number of observations in each group k = total number of parameters
After calculating this, we will compare this value to the F‐table to see if we reject the
13 null hypothesis that there is no difference between S1 and S2. If we reject it, then
there is a statistically significant difference between these two. This tells us that
13 http://en.wikipedia.org/wiki/Chow_test. Used to acquire equation
33 there was a change in this PointPct‐1 in the two time periods. The calculations for this test are given in Exhibit 11
Exhibit 11: Chow Test for Exhibit 10 Regressions Sc = .547 S1 = .602 N1 = 145 S2 = .457 N2 = 150 k = 2
Are null hypothesis is S1 = S2.
Next, we apply these numbers to the following equation
(Sc – (S1 + S2)) / k , (S1 + S2) / (N1 + N2 – 2K)
= (.547- (.602 +.457)) / 2 (.602 + .457) / (150+145-4)
= (.547 – 1.059) / 2 1.059 / 291
= -70.34 F(2, 291) ≈ 3
Since 70.34 > 3, reject the null hypothesis Regressions used in this model can be found in Appendix B (Equations 10.1, 10.2, 10.3) The critical F value is taken from the F table14
This exhibit shows that the beta values for Pointct‐1 are statistically different. Thus,
there is a difference in the effect that last year’s point percentage has on this year’s in the two time periods.
14 Stock and Watson, page 357
34
Determinants of Competitive Balance
The next step in this chapter is to look at the determinants of competitive
balance. The basic regression model is;
PointPct = α + β1Payroll + β2TFA1 + β3TFA2 + β4TFA3 + ε
Where PointPct = Point percentage Payroll = Payroll TFA1 = top‐end free agent TFA2 = medium‐end free agent TFA3 = low end free agent
PAYROLL REGRESSIONS
The first set of regressions will look just at payroll and its effect on point percentage.
The results are given in Exhibit 12.
Exhibit 12: Payroll Regressions Dependent Variable: PointPct
Mean Sample Dependent Size Variable C Payroll Payroll2 logPayroll R2 .003 Eq. 148 0.5271 .41 12.1a (18.64) (5.645) 0.17 .002 Eq. 150 0.5576 .456 12.1b (13.95) (3.17) 0.063 .13 Eq. 148 0.5271 .059 12.2a (.815) (6.4) 0.219 .098 Eq. 150 0.5576 .187 12.2b (1.67) (3.33) 0.07 .012 -.0001 Eq. 148 0.5271 .237 12.3a (4.14) (4.29) (-3.256) 0.23 .008 0 Eq. 150 0.5576 0.336 12.3b (.167) (1.36) (-.98) 0.069 Source: Appendix B
35 Equations 12.1a (pre lockout) and 12.1b (post lockout) demonstrate the relationship if a linear model is used. The R2 value for the seasons before the lockout is higher than for the seasons after. This means that payroll was a better predictor of point percentage before the lockout. To see if the beta values are different, a Chow test will be used. The results from the Chow test are given in
Exhibit 13.
Exhibit 13: Chow Test for Equations 12.1 Sc = .00292 S1 = ..00309 N1 = 148 S2 = ..00224 N2 = 150 k = 2
Are null hypothesis is S1 = S2.
Next, we apply these numbers to the following equation
(Sc – (S1 + S2)) / k , (S1 + S2) / (N1 + N2 – 2K)
= (.00292- (.00309 +.00224)) / 2 (.00309 +.00224) / (150+148-4)
= (.00292 – .00533) / 2 .00533 / 294
= -66.47 F(2, 294) ≈ 3
Since 66.47 > 3, reject the null hypothesis Regressions used in this model can be found in Appendix C (Equations 12.1a,b,c) The critical F value is taken from the F table15
15 Stock and Watson, page 357
36 Exhibit 13 demonstrates that the beta values in Equations 12.1a and 12.1b are
different. This tells us that the addition of payroll had a bigger effect on team
performance in the seasons before the lockout.
The next two equations look to see if a logarithmic relationship fits better
with the two variables. Again, the model that is used is:
PointPct = α + β1LogPayroll
This will tell us if there are diminishing returns to payroll. Equation 12.2a includes
the seasons before the lockout. This regression gives beta value for logPayroll of
.13. This means that a 1% increase in payroll leads to an increase in point
percentage of .13. These results suggest that if a team were to raise payroll by 10%,
which is a likely scenario, then they would increase their point percentage by 1.3.
Point percentage cannot be more than 1, so this is impossible. The same is true for
equation 12.2b. This tells us that a logarithmic relationship is not an accurate
portrayal of the true story.
The final two equations experiment with a quadratic relationship. Again, this model is:
2 PointPct = α + β1Payroll + β2Payroll
The hypothesis is that there are diminishing returns to the amount of money that is
spent on payroll. Equation 12.3a demonstrates that, for every million dollars that a
team spends on payroll, the point percentage will go up. However, with each additional million, it will increase by a lesser amount. This means that the effect of increasing payroll from $25 million to $30 million will have a bigger effect on winning percentage than an increase from $40 million to $45 million, prior to the
37 lockout. Although this effect exists, it does not seem to be large. The R2 value demonstrates that this is the most accurate relationship (in the pre‐lockout era) of the ones explored. For equation 12.3b (post lockout), the same effect does not exist.
This could suggest that in the seasons after the lockout, the impact of payroll is negligible. This is supported by the low R2 value of equation 12.3b.
Free Agency Impact
The next set of regressions includes free agent singings in the model. All of the following equations are only for the seasons after the lockout. This is because the data for free agents signed could not be found for the seasons before the lockout.
The free agency signings are split into three groups, based on their salary. Low‐end players (TFA3) will consist of players signed for under one million dollars. Medium‐ end players (TFA2) will be players making between one and four million dollars per year. The high‐end players (TFA3) will be the players making over four million dollars. The system is a little different for goalies. Starters will be in the top group, back‐ups will be in the medium group, and any other goalie will be in the third group. In doing this, we can see if there is a difference as to which players a team targets in free agency. The regression results are reported in Exhibit 14.
38 Exhibit 14: Model Statement Regressions Dependent Variable: PointPct
Mean Sample Dependent 2 Size Variable C Payroll TFA1 TFA2 TFA3 R
Eq. 150 0.5576 .492 0.002 .015 -0.002 -0.011 14.1 (15.42) (3.15) (2.29) (-0527) (-3.499) 0.195
Eq. 150 0.5576 .459 .0016 .016 14.2 (14.26) (2.28) (2.38) 0.098
Eq. 150 0.5576 .488 .0022 .015 -.011 14.3 (15.57 (3.13) (2.26) (-4.15) 0.193 Source: Appendix C
The regressions in these groups all give similar results for certain variables.
Equation 14.1 shows that the variables Payroll and TFA1 have a positive and
significant effect on point percentage. The variable TFA3 has a negative effect. This
tells us that spending more money on payroll and the addition of top‐end free
agents all increase a team’s point percentage (.002 for every million and .015 for
each free agent signed). Also, signing a low‐end free agent will decrease a team’s
point percentage by .011. This equation has an R2 value of .195, suggesting that
there are many more variables that have an effect on a team’s point percentage. The
next equation, Eq. 14.2, takes out the effect of signing low‐end and medium‐end
players, which results in an even lower R2 value. This does not change the effects of
payroll and TFA1. Equation 14.3 gives only the variables that are statistically
significant, and is still similar to the equation containing all of the variables. It also
has a similar R2 value, .193, to equation 14.1.
In general, these regressions tell us that the addition of a high‐end free agent
will increase the number of wins by about 1.2, which is equivalent to about 2.46
39 points in the standings. If a team were to have signed just 2 top‐end players, this
would increase their win total by 2 or 3 games, which is the equivalent of about 5
points in the standings. This may not seem like it is worth it, but 5 points is a big
difference for some teams. Exhibit 15 demonstrates this difference.
Exhibit 15: The Effect of Five Points on Playoff Teams Season 09‐10 08‐09 07‐08 06‐07 05‐06
3 4 5 4 3 Teams within 5 points of making playoffs
Teams within 5 points of increasing playoff 10 8 9 8 8 seed
From this exhibit, it is clear that 5 points in the standings can make a big difference
for some teams. There are only 16 teams that make the playoffs. In the past 5 years,
at least half of the playoff teams were within five points of increasing their playoff
seeds. This shows how close these teams are together. Effective free agent signing
can put a team ahead of the pack in these close scenarios. These regressions also
show the negative effects of signing a low‐end free agent. These low‐end players are
often players who do not play in the NHL. Signing these players uses up resources
that could be used to sign top‐end free agents or help in other ways.
40 Chapter 5 Interpretation of Findings and Conclusion
In this chapter, we interpret the findings on the measures and determinants
of competitive balance and conclude that the League’s competitive balance
improved as a result of the 2005 CBA. The impact of these changes in the CBA
allowed all teams to sign players of talent, leading to an increased in competitive
balance. We end this chapter with limitations of this study and suggestions for
future research
Interpretation of Findings
Measur es of Competitive Balance
Three different measures of a teams point percentage indicate that there has been a decrease in the dispersion of point percentage across the League.
The next two measures use the distribution of wins across teams. The H‐H index reduced, suggesting an increase in competitive balance. The quintile approach shows that the distribution of wins became more uniform as a result of the 2005
CBA. The next two measures used the playoffs as a measure. The number of playoff
games showed no change in the seasons after the lockout. However, the H‐H index
measuring the number of playoff appearances greatly decreased after the lockout.
This means different teams are making the playoffs each year. The final measure is
the effect that last year’s performance had on this years. The results show that this
effect is lower in the seasons after the lockout, suggesting more change in the
rankings from year to year.
41 In total, eight different measures were used as alternative measures of
competitive balance. Seven of these eight showed increases in competitive balance.
Some of these measures took a couple of seasons after the lockout to show a change.
This lag in the results is likely due to the amount of time it took for these effects to take place. The changes that were made were not instantaneous changes. It would take time for weaker teams to build a better team through free agency and spending money received through the revenue sharing plan. Also, even though some of the changes demonstrated are not substantial, the League is better than where it was before the lockout. This means the NHL has taken a step in the right direction with the 2005 CBA and, as a result, the competitive balance of the League has improved.
Determinants of Competitive Balance
The first regressions experimented with the effect that payroll has on a team’s point percentage. In each case looked at, payroll had less of an effect on team’s point percentage in the seasons after the lockout. This means that after the lockout, the effect of spending more money has less effect on the success of the team. Before the lockout, teams would spend lots of money to make the best team.
With the institution of the salary cap, this is no longer possible. By limiting the payroll of teams, and the inclusion of new revenue sharing plan, the dispersion of
payroll has decreased.
The next set of regressions include the free agent variables as well as payroll.
All of these regressions are from after the lockout. Here, we see a positive effect of
signing a top‐end free agent, and a negative effect of signing a low‐end free agent.
42 Although the impact of these signings on performance seem low, it can be the
difference between making the playoffs and missing them. In the eyes of fans and
managers, this is a huge difference.
From these two sets of regressions, we can draw some conclusions on the
determinants of competitive balance that this paper tested. For one, payroll does
have a positive effect on a team’s point percentage. This effect has decreased since
the 2005 CBA. Teams that go out and spend money will increase their team’s
performance, but can only do so by so much. The spread of payroll across teams is
certainly one of the controlling factors of competitive balance. Also, the signing of
top‐end free agents has shown to increase a team’s point percentage. Free agents
can sign with any team, so long as they have the cap room to do so. This gives
weaker teams a chance to sign some top‐end free agents to increase their team’s performance. Free agency signings is also a key determinant of competitive balance.
Conclusion
We conclude that competitive balance has increased as a result of the new
NHL Collective Bargaining Agreement. These changes are due to the institution of
the salary cap, changes in free agency rules, and the new revenue sharing plan. The
salary cap has shown to bring teams’ payrolls closer together which has lowered the
effect that payroll has on performance. The changes in free agency rules have helped teams to sign younger and better players. This can help a team to increase its performance over the course of a couple years, as long as they are signing top‐ end players. Finally, the revenue sharing plan has helped increase the competitive
43 balance of the League. With poorer teams getting money, they are able to spend it
on signing free agents and bring up their payroll so it is closer to the rest of the
League. These three changes are some of the driving forces that have led to an
increase in the competitive balance of the NHL.
The findings of this paper are important for many reasons. First, one of the
main goals of the 2005 CBA was to increase the competitive balance of the League.
That is why rules were introduced that would help weaker and poorer teams
perform better and put limits on the stronger and richer teams. The findings of this
paper suggest that the League has accomplished its goal. As the competitiveness of
the League has increased, so too have the ratings. Every playoff year since the lockout, there have been reports that television ratings have reached new heights.
This is a great sign for the NHL as they look to continue its success into the future.
This also important as the 2005 CBA agreement will expire after the 2011‐
2012 season, about a year and three months away. As this time gets closer, the NHL and NHLPA will look at the current state of the NHL to see if the changes that were made in the last CBA worked or not. This type of research will help them answer these questions. Also, it is possible that they can change the rules even more if they
want more competitive balance in the NHL. Examples for this would be to lower the
age of unrestricted free agents even more or increase the amount of revenue
sharing.
There are a few limitations to this study. For one, free agency signings was
not acquired for the years before the lockout. This information would have been
helpful for comparison to see if the changes made any difference or not. Also, there
44 was no real way to look at the effects that revenue sharing had for the teams that received it, as there was no apparent way to test this. As a result, this paper just assumed that teams were able to spend the money they received to increase payroll and possibly sign free agents.
In conclusion, no matter the measure used, competitive balanced increased as a result of the 2005 CBA. The impact of the salary cap, revenue sharing, and free agency rule changes equalized a team’s capacity to sign players of talent. This prevented stronger teams from acquiring all the talent and made the weaker teams more competitive, which is why the overall competitive balance improved. Going forward, these changes in the CBA should be preserved if the League is going to continue to improve its competitive balance.
45 Bibliography
About.com, Hockey, 2011.
Andreff, Wladimir & Szymanski, Stefan. Handbook on the Economics of Sport. Northampton, MA: Edward Elgar Publishing, 2006.
Berri David & Jewell, R. Todd, 2004. "Wage inequality and firm performance: Professional basketball's natural experiment," Atlantic Economic Journal, International Atlantic Economic Society, vol. 32(2), pages 130‐139.
Bissonnette, Zac. “Is the NHL on its way to Reviving Hockey,” Bloggin Stocks. December 3rd, 2007.
Castillo, Jorge. “NBA Free Agency: Salary Cap Exceptions Explained.” The Washington Post. July 3, 2010.
Chang Yand‐Ming, Sanders S., 2009. “Pool revenue sharing, team investments, and competitive balance in professional sports: A theoretical analysis.” Journal of Sports Economics, 10 (4), pp. 409‐428.
Dietl, Helmut & Lang, Markus & Rathke, Alexander 2007. "The Effect of Salary Caps in Professional Team Sports on Social Welfare," Working Papers 0072, University of Zurich, Institute for Strategy and Business Economics (ISU).
Fenn, Aiu & Allmen, Peter & Brook, Stacey & Preissing, Thomas, 2005. "The Influence of Structural Changes and International Players on Competitive Balance in the NHL," Atlantic Economic Journal, International Atlantic Economic Society, vol. 33(2), pages 215‐224, June.
Fishman, Peter, 2003. “Competitive Balance and Free Agency in Major League Baseball,” American Economist
Financial Times Lexicon, 2011. Financial Times LTD.
Forrest, David & Simmons, Robert, 2000. “The Relation Between Pay and Performance: Team Salaries and Playing Success From a Comparative Perspective,” University of Salford, Centre for Sports Economics.
Fort, Rodney & Lee, Young Hoon, 2007. "Structural Change, Competitive Balance, And The Rest Of The Major Leagues," Economic Inquiry, Western Economic Association International, vol. 45(3), pages 519‐532, 07.
46
The Hockey Nut, 2011.
Kesenne, Stefan, 2000. "The Impact of Salary Caps in Professional Team Sports," Scottish Journal of Political Economy, Scottish Economic Society, vol. 47(4), pages 422‐30, September.
Leeds, Michael and Allmen, Peter von. The Economics of Sports. Boston, MA: Pearson Education, Inc., 2002. Pages 252‐258
“MLB Collective Bargaining Agreement.” Major League Baseball Players Association, 2011.
Neale, W, 1964. “The Peculiar Economics of Professional Sports,” Quarterly Journal of Economics.
“NFL Collective Bargaining Agreement.” NFL Players Associations, 2011.
The National Hockey League Website, 2011.
“NHL Payrolls.” Hockey Zone Plus, 1999‐2010.
NHL Numbers, 2011.
The Official Website of the National Hockey League, 2011.
Owen, P.Dorian & Ryan, Michael and Weatherson, Clayton, 2006. “Measuring Competitive Balance in Professional Team Sports Using the Herfindahl‐Hirschman Index,” University of Otago Economics Discussion Paper.
Richardson, David H. 2000. "Pay, Performance, and Competitive Balance in the National Hockey League," Eastern Economic Journal, Eastern Economic Association, vol. 26(4), pages 393‐417, Fall.
Sanderson, Allen R. & Siegfried, John J, 2003. "Thinking About Competitive Balance," Working Papers 0318, Department of Economics, Vanderbilt University.
Sommers, Paul & Barriger, Andrew & Sharpe, John & Sullivan, Brendan, 2004. "Has a Salary Cap in the NFL Improved Competitive Balance?," Middlebury College Working Paper Series 0402, Middlebury College, Department of Economics.
Sports City, 2006‐2010.
47
Staudohar, Paul. “The Hockey Lockout of 2004‐2005.” All Business, 1999‐2011.
Stock, James and Watson, Mark. Introduction to Econometrics. Boston, MA: Pearson Education, Inc, 2008.
Szymanski, Stefan & Kesenne, Stefan, 2004. "Competitive balance and gate revenue sharing in team sports," Journal of Industrial Economics, Blackwell Publishing, vol. 52(1), pages 165‐177, 03.
Vrooman, John, 2009. "Theory of the Perfect Game: Competitive Balance in Monopoly Sports Leagues," Review of Industrial Organization, Springer, vol. 34(1), pages 5‐44, February.
48 Appendix A
Equation 10.1 PointPct = α + β1PointPct-1
Dependent Variable: POINTPCT Method: Panel Least Squares Date: 02/15/11 Time: 12:09 Sample: 2000 2004 Periods included: 5 Cross-sections included: 30 Total panel (unbalanced) observations: 145
Variable Coefficient Std. Error t-Statistic Prob.
POINTPCT-1 0.602251 0.062896 9.575339 0.0000 C 0.216400 0.033374 6.484029 0.0000
R-squared 0.390678 Mean dependent var 0.530517 Adjusted R-squared 0.386417 S.D. dependent var 0.094366 S.E. of regression 0.073919 Akaike info criterion -2.358010 Sum squared resid 0.781345 Schwarz criterion -2.316951 Log likelihood 172.9557 Hannan-Quinn criter. -2.341326 F-statistic 91.68712 Durbin-Watson stat 2.363822 Prob(F-statistic) 0.000000
Equation 10.2 PointPct = α + β1PointPct-1
Dependent Variable: POINTPCT Method: Panel Least Squares Date: 03/04/11 Time: 18:10 Sample: 2007 2011 Periods included: 5 Cross-sections included: 30 Total panel (balanced) observations: 150
Variable Coefficient Std. Error t-Statistic Prob.
PREVPOINTPCT 0.457687 0.067544 6.776148 0.0000 C 0.302678 0.038097 7.944943 0.0000
R-squared 0.236784 Mean dependent var 0.557893 Adjusted R-squared 0.231627 S.D. dependent var 0.080041 S.E. of regression 0.070162 Akaike info criterion -2.462787 Sum squared resid 0.728553 Schwarz criterion -2.422645 Log likelihood 186.7090 Hannan-Quinn criter. -2.446478 F-statistic 45.91618 Durbin-Watson stat 1.944366 Prob(F-statistic) 0.000000
49 Equation 10.3 PointPct = α + β1PointPct-1
Dependent Variable: POINTPCT Method: Panel Least Squares Date: 03/04/11 Time: 18:07 Sample: 2000 2011 Periods included: 10 Cross-sections included: 30 Total panel (unbalanced) observations: 295
Variable Coefficient Std. Error t-Statistic Prob.
PREVPOINTPCT 0.547116 0.045155 12.11631 0.0000 C 0.249049 0.024739 10.06699 0.0000
R-squared 0.333796 Mean dependent var 0.544437 Adjusted R-squared 0.331522 S.D. dependent var 0.088298 S.E. of regression 0.072192 Akaike info criterion -2.412205 Sum squared resid 1.527045 Schwarz criterion -2.387208 Log likelihood 357.8002 Hannan-Quinn criter. -2.402195 F-statistic 146.8050 Durbin-Watson stat 2.085300 Prob(F-statistic) 0.000000
50 Appendix B
Equation 12.1a PointPct = α + β1Payroll
Dependent Variable: POINTPCT Method: Panel Least Squares Sample: 2000 2004 Periods included: 5 Cross-sections included: 30 Total panel (unbalanced) observations: 148
Variable Coefficient Std. Error t-Statistic Prob.
PAYROLL 0.003088 0.000547 5.645322 0.0000 C 0.409911 0.021993 18.63825 0.0000
R-squared 0.179174 Mean dependent var 0.527101 Adjusted R-squared 0.173552 S.D. dependent var 0.097206 S.E. of regression 0.088369 Akaike info criterion -2.001160 Sum squared resid 1.140136 Schwarz criterion -1.960657 Log likelihood 150.0859 Hannan-Quinn criter. -1.984704 F-statistic 31.86966 Durbin-Watson stat 0.961259 Prob(F-statistic) 0.000000
Equation 12.1b PointPct = α + β1Payroll
Dependent Variable: POINTPCT Method: Panel Least Squares Sample: 2006 2010 Periods included: 5 Cross-sections included: 30 Total panel (balanced) observations: 150
Variable Coefficient Std. Error t-Statistic Prob.
PAYROLL 0.002238 0.000706 3.170579 0.0018 C 0.456162 0.032703 13.94847 0.0000
R-squared 0.063603 Mean dependent var 0.557620 Adjusted R-squared 0.057276 S.D. dependent var 0.085098 S.E. of regression 0.082625 Akaike info criterion -2.135756 Sum squared resid 1.010389 Schwarz criterion -2.095614 Log likelihood 162.1817 Hannan-Quinn criter. -2.119448 F-statistic 10.05257 Durbin-Watson stat 1.030052 Prob(F-statistic) 0.001849
51 Equation 12.1c PointPct = α + β1Payroll
Dependent Variable: POINTPCT Method: Panel Least Squares Sample: 2000 2010 Periods included: 10 Cross-sections included: 30 Total panel (unbalanced) observations: 298
Variable Coefficient Std. Error t-Statistic Prob.
PAYROLL 0.002919 0.000408 7.151551 0.0000 C 0.420832 0.017714 23.75702 0.0000
R-squared 0.147330 Mean dependent var 0.542463 Adjusted R-squared 0.144449 S.D. dependent var 0.092431 S.E. of regression 0.085495 Akaike info criterion -2.074030 Sum squared resid 2.163578 Schwarz criterion -2.049218 Log likelihood 311.0305 Hannan-Quinn criter. -2.064098 F-statistic 51.14469 Durbin-Watson stat 1.027127 Prob(F-statistic) 0.000000
52
Equation 12.2a PointPct = α + β1logPayroll
Dependent Variable: POINTPCT Method: Panel Least Squares Sample: 2000 2004 Periods included: 5 Cross-sections included: 30 Total panel (unbalanced) observations: 148
Variable Coefficient Std. Error t-Statistic Prob.
LOGPAYROLL 0.130662 0.020413 6.401071 0.0000 C 0.059791 0.073348 0.815165 0.4163
R-squared 0.219142 Mean dependent var 0.527101 Adjusted R-squared 0.213793 S.D. dependent var 0.097206 S.E. of regression 0.086191 Akaike info criterion -2.051077 Sum squared resid 1.084621 Schwarz criterion -2.010574 Log likelihood 153.7797 Hannan-Quinn criter. -2.034621 F-statistic 40.97371 Durbin-Watson stat 0.997904 Prob(F-statistic) 0.000000
Equation 12.2b PointPct = α + β1logPayroll
Dependent Variable: POINTPCT Method: Panel Least Squares Sample: 2006 2010 Periods included: 5 Cross-sections included: 30 Total panel (balanced) observations: 150
Variable Coefficient Std. Error t-Statistic Prob.
LOGPAYROLL 0.097933 0.029396 3.331552 0.0011 C 0.186484 0.111603 1.670956 0.0968
R-squared 0.069763 Mean dependent var 0.557620 Adjusted R-squared 0.063478 S.D. dependent var 0.085098 S.E. of regression 0.082353 Akaike info criterion -2.142357 Sum squared resid 1.003742 Schwarz criterion -2.102215 Log likelihood 162.6768 Hannan-Quinn criter. -2.126048 F-statistic 11.09924 Durbin-Watson stat 1.034178 Prob(F-statistic) 0.001091
53 Equation 12.3a 2 PointPct = α + β1Payroll + β2Payroll
Dependent Variable: POINTPCT Method: Panel Least Squares Date: 02/15/11 Time: 12:19 Sample: 2000 2004 Periods included: 5 Cross-sections included: 30 Total panel (unbalanced) observations: 148
Variable Coefficient Std. Error t-Statistic Prob.
PAYROLL 0.012117 0.002823 4.291831 0.0000 PAYROLLSQUARED -0.000105 3.23E-05 -3.255836 0.0014 C 0.237161 0.057176 4.147918 0.0001
R-squared 0.235094 Mean dependent var 0.527101 Adjusted R-squared 0.224544 S.D. dependent var 0.097206 S.E. of regression 0.085600 Akaike info criterion -2.058205 Sum squared resid 1.062463 Schwarz criterion -1.997450 Log likelihood 155.3071 Hannan-Quinn criter. -2.033520 F-statistic 22.28288 Durbin-Watson stat 1.011861 Prob(F-statistic) 0.000000
Equation 12.3b 2 PointPct = α + β1Payroll + β2Payroll
Dependent Variable: POINTPCT Method: Panel Least Squares Sample: 2006 2010 Periods included: 5 Cross-sections included: 30 Total panel (balanced) observations: 150
Variable Coefficient Std. Error t-Statistic Prob.
PAYROLL 0.007979 0.005858 1.362091 0.1753 PAYROLLSQUARED -6.52E-05 6.60E-05 -0.987323 0.3251 C 0.335812 0.126206 2.660815 0.0087
R-squared 0.069771 Mean dependent var 0.557620 Adjusted R-squared 0.057115 S.D. dependent var 0.085098 S.E. of regression 0.082632 Akaike info criterion -2.129032 Sum squared resid 1.003733 Schwarz criterion -2.068820 Log likelihood 162.6774 Hannan-Quinn criter. -2.104570 F-statistic 5.512833 Durbin-Watson stat 1.037726 Prob(F-statistic) 0.004913
54 Appendix C
Equation 14.1 PointPct = α + β1Payroll + β2TFA1 + β3TFA2 + β4TFA3
Dependent Variable: POINTPCT Method: Panel Least Squares Sample: 2006 2010 Periods included: 5 Cross-sections included: 30 Total panel (balanced) observations: 150
Variable Coefficient Std. Error t-Statistic Prob.
PAYROLL 0.002250 0.000713 3.157343 0.0019 TFA1 0.015082 0.006594 2.287026 0.0236 TFA2 -0.001989 0.003772 -0.527346 0.5988 TFA3 -0.010890 0.003112 -3.499738 0.0006 C 0.491602 0.031886 15.41733 0.0000
R-squared 0.195007 Mean dependent var 0.557620 Adjusted R-squared 0.172800 S.D. dependent var 0.085098 S.E. of regression 0.077397 Akaike info criterion -2.246962 Sum squared resid 0.868602 Schwarz criterion -2.146607 Log likelihood 173.5221 Hannan-Quinn criter. -2.206191 F-statistic 8.781423 Durbin-Watson stat 1.144001 Prob(F-statistic) 0.000002
Equation 14.2 PointPct = α + β1Payroll + β2TFA1
Dependent Variable: POINTPCT Method: Panel Least Squares Date: 02/22/11 Time: 10:52 Sample: 2006 2010 Periods included: 5 Cross-sections included: 30 Total panel (balanced) observations: 150
Variable Coefficient Std. Error t-Statistic Prob.
PAYROLL 0.001678 0.000733 2.288112 0.0236 TFA1 0.016443 0.006906 2.381135 0.0185 C 0.459817 0.032236 14.26413 0.0000
R-squared 0.098378 Mean dependent var 0.557620 Adjusted R-squared 0.086111 S.D. dependent var 0.085098 S.E. of regression 0.081352 Akaike info criterion -2.160268 Sum squared resid 0.972865 Schwarz criterion -2.100055 Log likelihood 165.0201 Hannan-Quinn criter. -2.135805 F-statistic 8.019779 Durbin-Watson stat 1.075192 Prob(F-statistic) 0.000495
55
Equation 14.3 PointPct = α + β1Payroll + β2TFA1 + β3TFA3
Dependent Variable: POINTPCT Method: Panel Least Squares Date: 02/22/11 Time: 10:53 Sample: 2006 2010 Periods included: 5 Cross-sections included: 30 Total panel (balanced) observations: 150
Variable Coefficient Std. Error t-Statistic Prob.
PAYROLL 0.002216 0.000708 3.130047 0.0021 TFA1 0.014860 0.006565 2.263612 0.0251 TFA3 -0.011602 0.002796 -4.148764 0.0001 C 0.488868 0.031384 15.57691 0.0000
R-squared 0.193463 Mean dependent var 0.557620 Adjusted R-squared 0.176890 S.D. dependent var 0.085098 S.E. of regression 0.077206 Akaike info criterion -2.258379 Sum squared resid 0.870268 Schwarz criterion -2.178096 Log likelihood 173.3784 Hannan-Quinn criter. -2.225763 F-statistic 11.67359 Durbin-Watson stat 1.144415 Prob(F-statistic) 0.000001
56